CN107977411A - Group recommending method, device, storage medium and server - Google Patents

Abstract

The invention discloses a group recommendation method, device, storage medium and server, and belongs to the technical field of the Internet. The method includes: for each user to be recommended, obtaining task preference information of the user to be recommended; among the candidate groups in an active state, according to the task preference information and the group characteristics of the candidate group, screening out the user who matches the user to be recommended The group to be predicted; obtain the probability that the user to be recommended joins each group in the group to be predicted; sort the groups contained in the group to be predicted according to the probability value, and obtain the first group recommendation list of the user to be recommended; based on The first group recommendation list performs group recommendation to the user to be recommended. The present invention combines the user's task preference and the group characteristics of the group to recommend the team for the user, which not only enriches the group recommendation method, makes the team recommendation more personalized, but also greatly improves the accuracy of the group recommendation. are also more likely to be active in groups.

Description

Group recommendation method, device, storage medium and server

technical field

The present invention relates to the technical field of the Internet, in particular to a group recommendation method, device, storage medium and server.

Background technique

As a kind of multiplayer online interactive game based on the network platform, online games have become a great pastime for people in their leisure time. Among them, most of the online games nowadays support users (also called players) to form teams, that is, a certain number of players can form a team. For each player in a team, they can form a team to do tasks, send gifts to each other, or chat with each other. Among them, teaming up to do tasks means that multiple players jointly complete a task. By completing the task, these players can not only obtain game props or points, experience the fun of the game, but also increase their competitive ability. And if the team is called a group, when the player is online, if it is possible to recommend groups that players can join, especially for players who have not joined any groups, it will be greatly improved. Improve user experience.

In related technologies, when recommending groups, firstly, the group is screened based on whether the player meets the joining conditions of a group. For example, if a group takes the player's points greater than M as the threshold, then for a player whose points are greater than M, the player meets the conditions for joining the group. And after filtering out the groups that meet the joining conditions for the player, by sorting these groups, a recommendation list about the groups can be obtained.

In the process of realizing the present invention, the inventors have found that the related technologies have at least the following problems:

Since the group recommendation is only carried out based on the single dimension of whether the user meets the joining conditions of the group, there are defects of single recommendation method and low recommendation accuracy.

Contents of the invention

In order to solve the problems in the prior art, embodiments of the present invention provide a group recommendation method, device, storage medium and server. Described technical scheme is as follows:

In the first aspect, a method for group recommendation is provided, the method comprising:

For each user to be recommended, obtain task preference information of the user to be recommended;

Among the candidate groups in an active state, according to the task preference information and the group characteristics of the candidate groups, filter out a group to be predicted that matches the user to be recommended;

Obtain the probability that the user to be recommended joins each group in the group to be predicted;

sorting the groups contained in the group to be predicted according to the probability value, and obtaining the first group recommendation list of the user to be recommended;

performing group recommendation to the user to be recommended based on the first group recommendation list.

In a second aspect, a group recommendation device is provided, and the device includes:

The first obtaining module is used to obtain task preference information of the user to be recommended for each user to be recommended;

A screening module, configured to, among candidate groups in an active state, filter out a group to be predicted that matches the user to be recommended according to the task preference information and the group characteristics of the candidate group;

The second obtaining module is used to obtain the probability that the user to be recommended joins each group in the group to be predicted;

A sorting module, configured to sort the groups contained in the group to be predicted according to the probability value, and obtain the first group recommendation list of the user to be recommended;

A recommendation module, configured to recommend groups to the user to be recommended based on the first group recommendation list.

In another embodiment, the groups in the first group recommendation list are arranged in descending order of probability values, and the recommendation module is configured to determine in the first group recommendation list the A specified group that has an associated relationship with the user to be recommended, the specified group is a group that includes a target user, and the target user is a user that has an associated relationship with the user to be recommended; according to the number of target users in the group, the In order from large to small, adjust the arrangement order of the groups contained in the specified group; place the specified group after sorting adjustment on the top of the first group recommendation list; based on the sorting adjustment process The first group recommendation list is used to make group recommendations to the user to be recommended.

In another embodiment, the recommendation module is configured to obtain, for each of the at least two groups, if the designated group includes at least two groups with the same number of target users. The degree of intimacy between the user to be recommended and the target user included in the group; sorting the at least two groups according to the degree of intimacy.

In another embodiment, the screening module is also used to screen users who are active within the first time period and have not joined any group to obtain the users to be recommended; or, screen the users who are active within the first time period The user who has joined the group but the group that has joined is an inactive group, obtains the user to be recommended; screens out the groups that are active within the first time period and has vacancies, and obtains the candidate group Group.

In another embodiment, the second acquiring module is configured to acquire the probability that the user to be recommended joins each of the groups to be predicted based on an established group prediction model.

In another embodiment, the device also includes:

The training module is used to use the first user who has joined the group and the group that has joined in the second time period as positive samples; for the second user who is active in the second time period and has not joined any group , randomly match at least one group for the second user in the positive sample to obtain a negative sample; obtain the first characteristic information of the positive sample, and obtain the second characteristic information of the negative sample ; performing model training based on the first feature information and the second feature information to obtain the group prediction model.

In another embodiment, the training module is further configured to, for each of the positive samples, obtain the first user profile information of the first user in the sample; obtain the group information contained in the sample The first group profile information of the group and the first member profile information of the member users in the group; characterizing the first user profile information, the first group profile information, and the first member profile information Information cross processing to obtain first cross feature information; performing continuous feature information in the first group profile information, the first user profile information, the first member profile information, and the first cross feature information Bucketing processing: performing one-bit effective encoding on the feature information after the bucketing process and the discrete feature information except the continuous feature information to obtain the feature information of the sample.

In another embodiment, the training module is further configured to, for each sample in the negative sample, obtain the second user profile information of the second user in the sample; obtain the group information contained in the sample The second group profile information of the group and the second member profile information of the member users in the group; characterizing the second user profile information, the second group profile information and the second member profile information Information cross processing to obtain second cross feature information; performing continuous feature information in the second group profile information, the second user profile information, the second member profile information, and the second cross feature information Bucketing processing: performing one-bit effective encoding on the feature information after the bucketing process and the discrete feature information except the continuous feature information to obtain the feature information of the sample.

In another embodiment, the recommendation module is configured to obtain a second group recommendation list of the user to be recommended, and the second group recommendation list is filtered based on whether the user to be recommended meets the group joining conditions Obtained; if there are duplicate groups in the first group recommendation list and the second group recommendation list, then filter out the duplicate groups in the second group recommendation list; based on The first group recommendation list and the second group recommendation list after filtering and deduplication perform group recommendation to the user to be recommended.

In another embodiment, the recommendation module is configured to send the first group recommendation list and the filtered and deduplicated second group recommendation list to the The terminal of the user to be recommended, so that the terminal lists the groups contained in the first group recommendation list and the second group recommendation list on the recommendation interface in a cross-sorted manner according to the arrangement order of the groups to show.

In another embodiment, the device also includes:

A collection module, configured to collect user logs of the user to be recommended, the user log recording the exposure of the recommended group and the operations performed by the user to be recommended on the recommended group;

An update module, configured to update the positive samples and negative samples used for the group prediction model training according to the user log;

The training module is used to optimize the group prediction model based on the updated positive samples and the updated negative samples.

In a third aspect, a storage medium is provided. At least one instruction, at least one program, code set or instruction set is stored in the storage medium, and the at least one instruction, the at least one program, the code set or the instruction The set is loaded and executed by the processor to implement the group recommendation method described in the first aspect above.

In a fourth aspect, a server for group recommendation is provided, the server includes a processor and a memory, at least one instruction, at least one program, a code set or an instruction set are stored in the memory, and the at least one instruction , the at least one program, the code set or instruction set is loaded and executed by the processor to implement the group recommendation method as described in the first aspect above.

The beneficial effects brought by the technical solution provided by the embodiments of the present invention are:

Combining the user's task preference and the group characteristics of the group to make group recommendations for users, it not only enriches the group recommendation methods, makes the group recommendation more personalized, but also greatly improves the accuracy of the group recommendation. are also more likely to be active in groups.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

FIG. 1 is a schematic diagram of an implementation environment involved in a group recommendation method provided by an embodiment of the present invention;

Fig. 2 is a schematic diagram of acquiring model features provided by an embodiment of the present invention;

Fig. 3 is a schematic diagram of an active portrait of a player provided by an embodiment of the present invention;

FIG. 4 is a schematic diagram of a mapping relationship between player types and team characteristics provided by an embodiment of the present invention;

Fig. 5 is a flowchart of a group recommendation method provided by an embodiment of the present invention;

FIG. 6 is a flow chart of a group recommendation method provided by an embodiment of the present invention;

Fig. 7 is a flow chart of a group recommendation method provided by an embodiment of the present invention;

Fig. 8 is a schematic diagram of a recommendation panel provided by an embodiment of the present invention;

FIG. 9 is a flowchart of a group optimization process provided by an embodiment of the present invention;

Fig. 10 is a flowchart of a group recommendation method provided by an embodiment of the present invention;

Fig. 11 is a schematic structural diagram of a group recommendation device provided by an embodiment of the present invention;

Fig. 12 is a schematic structural diagram of a group recommendation device provided by an embodiment of the present invention;

Fig. 13 is a schematic structural diagram of a group recommendation device provided by an embodiment of the present invention;

Fig. 14 is a schematic structural diagram of a server for group recommendation provided by an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the implementation manner of the present invention will be further described in detail below in conjunction with the accompanying drawings.

Before explaining and describing the embodiments of the present invention in detail, some terms involved in the embodiments of the present invention will be explained in detail.

Player: referred to as a user in the embodiment of the present invention.

Among them, players can also be called gamers, which is a term between the game industry and game participants. Broadly speaking, a player refers to a user who plays a game, that is, a person who participates in any form of gaming.

In particular, in role-playing games, players usually play controllable characters in the game world, and complete the game or the goals set by themselves by operating these controllable characters. In addition, some players can also serve as the protagonist or the key to the game plot in role-playing games.

To sum up, players are game experiencers, users, evaluators and consumers. According to differences in personality and preferences, different game types are preferred by different players.

Team: referred to as a group in the embodiment of the present invention.

Online games usually include multiplayer battles, that is, a certain number of players form a team to complete a task together. By completing the task, these users can not only obtain game props or points, experience the fun of the game, but also increase their competitive ability.

It should be noted that, in the embodiment of the present invention, a field for saving team characteristics is respectively set for each team. Among them, the characteristics of the team are used to refer to the category or characteristics of a team, and are also called group characteristics in the embodiment of the present invention. This embodiment of the present invention does not specifically limit it.

Active players: Refers to players whose online time is greater than a set threshold within a period of time.

For example, if the set time period is 1 day and the set threshold is 10 minutes as an example, if a player's online time in 1 day is longer than 10 minutes, then the player is an active player.

Active team: Whether a team is active in the example of the present invention can be measured from the following two aspects at least: (1), whether there are active players in a period of time; (2), whether the team points are increasing. A team is active if there are active players and team points are increasing for a period of time. Continuing to take the set time period as 1 month as an example, if a team has active players within 1 month and the team's points are increasing, the team will be regarded as an active team. Of course, whether a team is active can also be measured from other aspects, which is not specifically limited in this embodiment of the present invention.

Friends: a concept that exists in social applications or applications with social functions. In social applications or applications with social functions, each user corresponds to a user relationship chain (also called an address book). All of the user's friends are shown in the chain. Among them, the two parties who are mutual friends can exchange information, such as voice, video or text chat, file transfer, and check each other's dynamics. In the embodiment of the present invention, a user has an association relationship with another user, which means that the two are friends with each other.

Active portrait: Based on the player's online time, the ratio of the player's game time in various types of gameplay to the total online time as the dimension, construct an active portrait for each player.

In other words, the active portraits can be used to classify the players, that is, the active portraits show the task preference information of different players. Wherein, the task preference information may also be referred to as game play preference information in the embodiment of the present invention, which shows the game play that the player is keen on or likes.

Secondary retention rate/weekly retention rate: After joining a team, participating in a gameplay or an event, the proportion of players who are still active the next day/week after joining the team, participating in this gameplay or activity.

Fig. 1 shows the implementation environment involved in the group recommendation method provided by the embodiment of the present invention.

Referring to FIG. 1 , the implementation environment may include: a service server 101 (also called a service side), a data server 102 (also called a data side) and a terminal 103 .

Among them, through the interaction between the business server 101 and the data server 102, personalized team recommendation is implemented for the player in combination with the player's game play preference and the team's team characteristics.

In the embodiment of the present invention, the business server 101 and the data server 102 provide the team recommendation service to the terminal through the game application installed on the terminal 103, that is, after the business server 101 and the data server 102 obtain the team recommendation list about the player, The game application installed on the terminal 103 pushes the team recommendation list to the terminal 103, and then the terminal 103 can display the team recommendation list to the player through the installed game application.

Wherein, the terminal 103 may be a smart phone, a tablet computer, a desktop computer, etc., which is not specifically limited in this embodiment of the present invention. When the terminal 103 displays the team recommendation list to the player through the installed game application, it may specifically set up a recommendation panel dedicated to displaying the team recommendation list. The panel displays the recommended list of the team. The above is only an exemplary display manner of the team recommendation list, and the embodiment of the present invention does not specifically limit the display manner of the team recommendation list.

The first point to be explained is that the above-mentioned service server 101, data server 102 and terminal 103 can be connected through a wired network or a wireless network.

The second point to be explained is that, in the embodiment of the present invention, two servers, the business server 101 and the data server 102, are used to realize the above-mentioned personalized team recommendation. Scalably, the above-mentioned service server 101 and data server 102 can also be replaced by a separate server that integrates all the functions of the two, which is not specifically limited in this embodiment of the present invention.

In another embodiment, the embodiment of the present invention involves the screening of candidate sets, training sets and prediction sets, as well as the training of group prediction models (machine learning models) in order to implement group recommendations. Before explaining the group recommendation method provided by the embodiment of the present invention in detail, first explain the above concept and model training process.

candidate set

In the embodiment of the present invention, the data server will screen players and teams, and filter out inactive players and teams, because on the one hand, recommend teams to inactive players or recommend inactive players Active teams are of little significance. On the other hand, filtering out inactive players and teams can reduce the prediction pressure of the group prediction model.

In short, the candidate set is used to screen players and teams that are candidates for recommendation. That is, only the players in the candidate set are recommended for teams, and the teams recommended to these players are also from the candidate set.

Wherein, when screening the teams in the candidate set, it includes but is not limited to: screening designated teams that are active within the first time period to obtain candidate teams.

Taking the first time period as one month as an example, the data server can filter the specified teams that have been active in the last month, and use these teams as the above-mentioned candidate teams. Among them, the candidate team usually needs to meet the following conditions:

1. There are active players in the first period of time, for example, there are active players in the last month.

2. Team points are increasing.

It should be noted that a team can be determined as an active team through the above two conditions. However, the above two conditions alone cannot determine that the team can be used as a candidate team. Because the candidate team in the embodiment of the present invention may be recommended to the player in the future, if the number of members in the team is full or the team does not recruit members, it is meaningless to recommend the team to the player. Based on this, in The following condition 3 is also included in the selection of candidate teams.

3. The team has recruited members and there are vacancies.

To sum up, through the above-mentioned conditions, the data server can filter out the teams that are active and have vacancies, that is, the designated teams mentioned above, and add the selected designated teams to the candidate set as candidate teams.

Among them, when screening the players in the candidate set, it includes but is not limited to the following two methods:

Method 1: Screen the players who are active within the first period of time and have not joined any group, and obtain the players to be recommended.

This method is used to screen out active players who have not joined the team recently, and add these players to the candidate set as players to be recommended. In addition, trumpets can also be filtered out when screening players to be recommended. Wherein, the identification method of the trumpet may be low number of entries, low account level, long chat time, etc., which is not specifically limited in this embodiment of the present invention.

Method 2: Screen the users who have joined the group within the first time period but the joined group is an inactive group, and obtain users to be recommended;

This method is used to filter out players who have recently joined a team, but the team they joined is not active.

In addition, the data server may periodically screen the candidate set, for example, once a week to screen candidate teams and players to be recommended within the window range of the last 30 days, which is not specifically limited in this embodiment of the present invention.

Training set

In the embodiment of the present invention, the training set is used for training the group prediction model. Wherein, the group prediction model is a machine learning model, such as an LR (Linear Regression, linear regression) model, a GBDT (GradientBoosting Decision Tree, gradient boosting decision tree) model or a random forest model, which is not described in detail in the embodiment of the present invention. limited. The embodiment of the present invention only uses the LR model as an example for illustration.

In the embodiment of the present invention, when training the LR model, the training set is constructed based on whether the player joins the team. In short, the players who have joined the team in the near future and the corresponding joined team are taken as the positive samples in the training set, and the players who are active recently and have not joined the team are selected, randomly matched with the teams in the positive sample, and then trained Concentrated negative samples.

That is, the data server uses the first user who has joined the team within the second time period and the team that has joined as positive samples. For the second user who is active within the second time period and has not joined any team, the second user is randomly matched with at least one team in the positive sample to obtain a negative sample.

Wherein, the first time period may be the same as or different from the first time period, which is not specifically limited in this embodiment of the present invention, for example, the second time period may be within one month. The reason why the embodiment of the present invention selects recent data is that the timeliness of features is considered, so the training set is not aimed at the full amount of data.

The first point to be explained is that although the ultimate goal of the embodiment of the present invention is to predict whether a player will click on a recommended team, the goal of the current LR model is to predict whether a player will add a certain recommended team. But these two goals are positively correlated. Only when players are interested in the team first and click to apply for the team, can they finally join the team.

The second point that needs to be explained is that in some cases, although a player may have joined a team, the player is still not in a team because the current application has not been passed or the data update is delayed. state, and thus lead to some deviations in the construction method of the above-mentioned negative examples, but one thing to note is that, assuming that a player can only join one team in the near future, for an active player who has not joined any team recently, although In the embodiment of the present invention, multiple teams are randomly matched for the player, that is, the player corresponds to multiple negative examples, but only one of the negative examples has a wrong label, and the remaining negative examples are all correct, so A large number of correctly labeled negative samples can dilute this wrongly labeled negative sample, that is, the deviation of the above-mentioned negative samples can be completely ignored.

In the embodiment of the present invention, after obtaining the training set, the data server needs to extract the model features first, that is, it needs to first obtain the first feature information of the positive sample and the second feature information of the negative sample, and then based on the first The first feature information and the second feature information are used for model training to obtain an LR model.

Among them, obtaining the first feature information of positive samples includes but not limited to the following steps:

(1) For each sample in the positive sample, obtain the profile information of the first player of the first player in the sample.

Among them, a sample is a matching pair of a player and a team. In the embodiment of the present invention, the player included in a positive sample is collectively referred to as the first player. The first player profile information may include basic player information such as the player's registration time, game duration, VIP (Very Important People, VIP) level, gender, and number of entries, which is not specifically limited in this embodiment of the present invention.

(2) Obtain the brief introduction information of the first team of the team in the sample and the brief introduction information of the first member of each member player in the team.

Wherein, the profile information of the team included in a positive sample is collectively referred to as the first team profile information, and the profile information of each member player in the team is collectively referred to as the first member profile information.

In this embodiment of the present invention, the first team profile information may include team basic information such as team registration time, team points, number of members, etc., which is not specifically limited in this embodiment of the present invention. The first member profile information usually includes the average value of the basic information of the members in the team, such as the average game duration and the average number of competitions of the members in the team, which are not specifically limited in the embodiment of the present invention. Wherein, the basic information of members may include registration time, game duration, VIP level, gender, number of competitions, etc. of each member, which is also not specifically limited in this embodiment of the present invention. It should be noted that the basic information of the member refers to the basic information of the remaining member users except the first user.

(3) Perform feature information intersection processing on the profile information of the first player, profile information of the first team, and profile information of the first member in the sample to obtain the first profile information of the cross.

As shown in Figure 2, in order to increase the expressive ability of the LR model, the data server also performs cross-processing of feature information. Wherein, the feature information cross processing is to perform cross merge processing on different feature information. Taking one feature information as female gender and another feature information as VIP level as VIP4 as an example, after the two feature information are cross-processed, a cross feature information such as female+VIP4 can be obtained.

(4) The first team profile information, the first player profile information, the first member profile information, and the continuous feature information in the first cross feature information are bucketed.

In the embodiment of the present invention, in order to improve the data management efficiency or process the noise data to make the model more stable, the continuous feature information in all the acquired feature information is also processed into buckets. Among them, continuous feature information usually refers to features whose value is a rational number but the number of feature values is indefinite. For example, the game duration feature, the feature value can range from 0 to positive infinity. The bucketing process is to divide a large amount of characteristic information into a more fine-grained data range. Each bucket only focuses on part of the characteristic information, which improves the efficiency of data management.

(5) Perform one-bit effective encoding on the feature information after the bucketing process and the discrete feature information except the continuous feature information to obtain the feature information of the sample.

Among them, one effective code specifically refers to one-hot code, which is used to digitize features for use in machine learning algorithms. One-hot encoding mainly uses N-bit state registers to encode N states, each state has its own independent register bit, and only one bit is valid at any time.

In actual machine learning tasks, some features are not continuous values, but may be some categorical values, such as gender can be divided into "male" and "female". For such features, it is usually necessary to digitize them, such as the following three feature attributes:

gender: ["male", "female"]

Region: ["Europe", "US", "Asia"]

Browser: ["Firefox", "Chrome", "Safari", "Internet Explorer"]

Assuming a sample is ["male", "US", "Internet Explorer"], if the features of this sample are serialized into [0,1,3] when digitizing, such feature processing cannot be directly put into the machine In the learning algorithm, it is necessary to introduce one-hot encoding. And if one-hot encoding is used to encode the sample "["male", "US", "InternetExplorer"]", then "male" can correspond to [1, 0], and "US" can correspond to [0, 1, 0] ], "Internet Explorer" can correspond to [0,0,0,1], then the digitized result of the full version of the sample is: [1,0,0,1,0,0,0,0,1].

To sum up, each sample in the positive sample is processed in the above manner, and the first feature information of the positive sample can be obtained. Similarly, when obtaining the second characteristic information of the negative sample, it includes:

First, for each sample in the negative sample, obtain the second player profile information of the second player in the sample, the second team profile information of the team in the sample, and the second member profile information of each member player in the team.

Afterwards, the profile information of the second player, the profile information of the second team, and the profile information of the second member are interleaved with the feature information to obtain the second cross feature information, and the profile information of the second team, the profile information of the second player, and the profile information of the second member are obtained. The profile information and the continuous feature information in the second cross feature information are processed in buckets, and finally the feature information after the bucketing process and the discrete feature information except the continuous feature information are effectively coded by one bit to obtain the feature of the sample information.

To sum up, each sample in the negative sample is processed in the above manner, and the second feature information of the negative sample can be obtained.

In summary, for each sample in the training set, the embodiment of the present invention will execute the above-mentioned process to obtain model features for training the team recommendation model. Wherein, the model features include team feature information about teams, player feature information about players, and cross feature information after feature information cross processing.

prediction set

In the embodiment of the present invention, after the candidate set is screened out, players and teams need to be matched and combined to form a prediction set. The reason for this is that if each player predicts all teams, the amount of data may be around 100 billion, which is too large, and team recommendation is difficult to achieve. For example, assuming that the candidate set includes 30 million players to be recommended as teams and 30,000 teams, if each player recommends a team from the 30,000 teams, the amount of data to be processed is too large.

To solve this problem, the embodiment of the present invention can classify players according to active portraits, that is, divide them into different game play preferences or player categories. Since each team has a field of team characteristics, that is, each team includes a group feature, a mapping is made for player category/game play preference, team feature/group feature. For example, the mapping relationship between player category/game play preference and team characteristics/group characteristics can be preset and saved. In this way, for players with different game play preferences, they can select from the matching teams to be predicted when recommending teams. For example, for players who are keen on competition, a competition team can be selected as the team to be predicted.

Continuing to take the above example as an example, if a player is matched with 300 teams to be predicted through the above mapping relationship, then the team will be selected from the 300 teams to be predicted when recommending a team.

For example, referring to FIG. 3 , players are divided into 6 categories, namely mode 1, mode 2, mode 3, mode 4, trumpet and core. Among them, modes 1 to 4 usually refer to different gameplays of the same game. For example, modes 1 to 4 can be several modes such as extreme speed mode, pet mode, dance mode, competition mode, leisure mode, and prop mode, which are not specifically limited in this embodiment of the present invention. For a player, when judging which of the above categories the player belongs to, that is, determining a player's task preference, it is based on the player's active portrait.

As shown in Figure 3, for player 1, because his active portrait shows that he has the longest game time in mode 4, player 1 belongs to mode 4; for player 3, because his active portrait shows Player 1 belongs to Mode 1 because he has the longest game time in Mode 1; Player 5 belongs to Mode 5 because his active profile shows that he has the longest game time in Mode 3 3. For player 6, since his active portrait shows that he has the longest game time in mode 2, player 1 belongs to mode 2.

As for player 2, since his active portrait shows that his online time is very short, and the distance from mode 1 to mode 4 is relatively long, player 2 is not keen on any kind of gameplay and belongs to the trumpet. For example, Player 2 may only be chatting online briefly.

As for player 4, since his active profile shows that he has been online for the longest time, and the distance from mode 1 to mode 4 is similar, player 4 belongs to the core player. For example, Player 4 may participate in each game play, and the game time in each game play is about the same.

Wherein, a schematic diagram of the mapping relationship between player categories and team characteristics may be shown in FIG. 4 . For example, a player whose player type is mode 1 is matched with a team whose team feature is casual mode, so when recommending a team for this type of player, it is also necessary to select from these teams. Players whose player type is mode 4 are matched with teams whose team characteristics are unlimited and game mode, so when recommending teams for such players, they can choose from the teams whose team characteristics are these two modes.

The above content has explained in detail the screening of candidate sets, training sets, and prediction sets, as well as the training process of the group prediction model. The team recommendation process provided by the embodiment of the present invention will be described in detail below in combination with the above content.

Fig. 5 is a flow chart of a team recommendation method provided by an embodiment of the present invention. Referring to Figure 5, the method flow provided by the embodiment of the present invention includes:

501. For each player to be recommended, the data server obtains the task preference information of the player to be recommended and, among the active candidate teams, screens out the ones that match the user to be recommended according to the task preference information and the group characteristics of the candidate teams. To be predicted team.

The recommended players, task preference information, group characteristics, candidate teams, and teams to be predicted have been introduced in detail above, so I won’t repeat them here.

For this step, among the active candidate teams, if the group characteristics of a candidate team match the task preference information, then the candidate team is a team to be predicted. For example, suppose that the task preference information is a competition mode, and if the group feature of a candidate team indicates that the candidate team is a competition team, then the two match each other.

502. Based on the established group prediction model, the data server obtains the probability of the player to be recommended joining each team in the team to be predicted, and sorts the teams included in the team to be predicted according to the probability value, and obtains the first team of the player to be recommended Recommended list.

In the embodiment of the present invention, the output of the group prediction model is the probability that the player to be recommended joins each team in the team to be predicted. Afterwards, the data server can sort the teams contained in the teams to be predicted according to the output probability values. For example, the teams in the teams to be predicted can be arranged in descending order of the probability values. This embodiment of the present invention does not specifically discuss this limited. It should be noted that if there are a large number of teams to be predicted, some teams may also be included in the first team recommendation list, such as only including the top 5% or 10% teams with the highest probability value, or only including the teams whose probability value is greater than a certain threshold team, which is not specifically limited in this embodiment of the present invention.

Afterwards, the business server and the data server can make group recommendations to players to be recommended based on the first team recommendation list. That is, the present invention also includes the following steps 503 and 504 .

503. After obtaining the recommendation request of the player to be recommended for the team, the service server pulls the first team recommendation list from the data server, and sends the first team recommendation list to the terminal of the user to be recommended.

In the embodiment of the present invention, a recommendation panel dedicated to displaying the recommended team can be set up. When a player who meets the aforementioned team recommendation conditions views the recommendation panel, the player will be regarded as a player to be recommended, and the team will be recommended for him. . It should be noted that when the player checks the recommendation panel, it will trigger the terminal to send a recommendation request for the team to the business server, and after receiving such a request, the business server will go to the data server to pull the first recommendation for the user to be recommended. Team recommendation list.

504. The terminal of the user to be recommended displays the recommendation list of the first team.

Wherein, if one page of the recommendation panel cannot display all the teams in the recommendation list of the first team, it can also be realized through multiple pages, which is not specifically limited in this embodiment of the present invention. For example, the first team recommendation list includes 20 teams, but only 10 teams can be displayed on one page at most, then the 20 teams can be displayed on 2 pages. For example, the first page displays the top 10 teams with probability values in descending order of probability values, and the second page displays the bottom 10 teams with probability values in descending order of probability values. clan.

In another embodiment, in order to improve the accuracy of team recommendation, the embodiment of the present invention further includes a process of cleaning the obtained first team recommendation list. That is, referring to FIG. 6 , the above step 502 may also be replaced by the following steps 505 and 506 .

505. The data server determines a designated team associated with the player to be recommended in the first team recommendation list, wherein the designated team is a team including the target player, and the target player is a player who has a friend relationship with the player to be recommended.

This step aims to find those teams that have friendship relationships with the players to be recommended, wherein the teams that have friendship relationships with the players to be recommended refer to those teams that are friends of the players to be recommended. After finding these teams, the embodiment of the present invention will Adjust the ranking order of the teams in the first team recommendation list obtained above, and give priority to displaying teams that have a friend relationship with the player to be recommended. For the specific adjustment process, please refer to step 506 below.

506. The data server adjusts the order of the teams included in the designated team in descending order of the number of target players, and places the designated team after sorting and adjustment at the top of the first team recommendation list.

In the embodiment of the present invention, for the above-mentioned specified teams, the data server will adjust the arrangement order of the teams contained in the specified teams according to the order of the number of friends from large to small, and then place the specified teams after sorting and adjustment in the first place. The top of the recommendation list of a team, to ensure that the team that has a friend relationship with the player to be recommended is displayed first.

Among them, when the data server adjusts the arrangement order of the teams contained in the designated team according to the order of the number of target players from large to small, it includes but is not limited to the following methods: if the specified team contains at least two players with the same number of friends For each of the at least two teams, obtain the intimacy between the player to be recommended and the friends contained in each team, and sort the at least two teams according to the obtained intimacy.

In the embodiment of the present invention, when the number of friends is the same, the higher the intimacy with the friends, the higher the priority for the corresponding team to display. In another embodiment, in addition to the above methods, the ranking of teams can also be guided according to the number of active friends. Among them, the concept of an active friend is similar to that of an active player, and refers to a friend whose online time is greater than a set threshold within a period of time.

Among them, the intimacy between players and friends can be measured from many aspects, such as whether the basic information between the two is similar, such as whether the game duration or number of entries or VIP levels are similar, whether the gender is the same, whether they have interacted, whether Jointly participated in activities, etc., which are not specifically limited in this embodiment of the present invention. That is, the degree of intimacy reflects the degree of intimacy between the player and the friend. Usually, the closer the friends are, the greater the influence they will have on them, so the priority of displaying the friends' teams is to take advantage of this feature.

To sum up, when sorting the teams of friends, the teams are sorted according to the number of friends. For the same number of friends, the teams are sorted according to the intimacy guide.

After the first team recommendation list obtained in the above step 502 is processed in the above steps 505 and 506, the final team recommendation list for team recommendation can be obtained, because this recommendation list is obtained based on the group prediction model , so it can also be called the model recommendation list.

In another embodiment, in addition to obtaining the above-mentioned model recommendation list in the embodiment of the present invention, when a player views the recommendation panel, the business server can pull the above-mentioned model recommendation list from the data server, and can also follow the original logic Get a list of teams (call it the second recommended list of teams). Wherein, the second team recommendation list is obtained by screening based on whether the players to be recommended meet the team joining conditions. That is, referring to FIG. 7 , on the basis of the scheme in FIG. 6 , this embodiment of the present invention further includes the following steps 507 to 510 .

507. After obtaining the recommendation request from the player to be recommended to the team, the service server obtains a second group recommendation list of the player to be recommended.

508. If there are duplicate teams in the first team recommendation list and the second team recommendation list after sorting and adjustment processing, the service server filters out the duplicate teams in the second team recommendation list.

The purpose of this step is to filter and rank the first team recommendation list and the second team recommendation list. If a team appears in both team recommendation lists, it will be deleted from the second team recommendation list.

Next, the service server can recommend a team to the player to be recommended based on the first team recommendation list and the second team recommendation list after filtering and deduplication.

It should be noted that the above steps of filtering and deduplication can be completed by the service server or the data server, which is not specifically limited in this embodiment of the present invention.

509. The service server sends the first team recommendation list after sorting and adjustment processing and the second team recommendation list after filtering and deduplication to the terminal of the player to be recommended.

510. The terminal of the user to be recommended cross-sorts and displays the recommended list of the first team after sorting and adjustment processing and the recommended list of the second team after filtering and deduplication according to the order of the teams.

The meaning of the order of the teams is that, for each team from the same team recommendation list, although the original order of each team may be broken due to the cross-sort display, on the whole, they are scattered on different pages of the recommendation panel The teams are still sorted in descending order of probability value.

In the embodiment of the present invention, as shown in FIG. 8 , the teams can be displayed in the following manner in the recommendation panel: N teams from the first team recommendation list and N teams from the second team recommendation list on each page List of clans. Taking the value of N as 5 as an example, teams a1 to a5 from the first team recommendation list can occupy positions 1, 3, 5, 7, and 9, while teams b1 to b5 from the second team recommendation list can occupy Occupies 2, 4, 6, 8, 10 positions. Of course, in addition to this display manner, other display manners may also be adopted, which is not specifically limited in this embodiment of the present invention.

In another embodiment, the solutions shown in steps 507 to 510 above can also be combined with the solution shown in FIG. 5 , and the specific process is similar to the combined solution shown in FIG. 7 , which will not be repeated here.

In another embodiment, no matter for the solutions shown in Figure 5, Figure 6 or Figure 7 above, after recommending teams to players, the embodiments of the present invention also include optimizing the group prediction model based on player logs (Also called update) process. As shown in Figure 9, this model optimization process mainly includes the following three steps.

901. The data server collects the player logs of the players to be recommended, and the player logs record the exposure of the recommended team and the operations performed by the players to be recommended on the recommended team.

Wherein, the player log is specifically collected by the player's terminal and reported to the service server, and then reported to the data server by the service server. Wherein, the collection of player logs may also be performed periodically, such as once a week, which is not specifically limited in this embodiment of the present invention. In the embodiment of the present invention, the operation performed on the recommended team may include clicking on a team (that is, the player applies for a team), the mouse pointer or cursor or the operation point hovering over a displayed team, the player applies for a team and passes, this The embodiment of the invention does not specifically limit this. Among them, if a team is displayed on the recommendation panel, it can be said that the team has been exposed. After a player applies for a team, usually the administrator of the team needs to approve it. Only after the approval is passed, the player can join the team.

902. The data server updates the positive samples and negative samples used for group prediction model training according to the player log.

This step can be used to correct the labels of the positive samples and negative samples mentioned above, so that the sample purity of positive samples and negative samples is higher; it can also expand the number of samples in positive samples or negative samples .

903. The data server optimizes the group prediction model based on the updated positive samples and the updated negative samples.

The initial group prediction model may not be very accurate, but after recommending teams based on this model, and then guiding the model optimization based on the collected player logs, the accuracy of the model can be made higher and higher. And the longer the time, the higher the accuracy of the obtained model.

In the following, the team recommendation process provided by the embodiment of the present invention will be briefly described in conjunction with the team recommendation process shown in FIG. 10 .

(a) The data side screens players to be recommended and candidate teams within one month every week as a candidate set.

(b) For each player to be recommended, the data side selects the team to be predicted from among the candidate teams.

(c) The data side uses the established group prediction model to obtain the probability of each player to be recommended joining each team in the team to be predicted.

(d) On the data side, each team in the predicted team is sorted based on the probability value, and the team that has a friend relationship with the player to be recommended is displayed first, and the model recommendation list is obtained.

(e) The player checks the recommendation panel, and the business side obtains the original recommendation list according to the original logic.

(f) The business side pulls the model recommendation list, and performs filtering and deduplication processing on the model recommendation list and the original recommendation list.

(g) The business side sends the model recommendation list and the original recommendation list after filtering and sorting to the terminal.

(h) The terminal displays the teams on at least one page of the recommendation panel according to the N teams from the model recommendation list and the N teams from the original recommendation list.

(i), the terminal counts the operations performed by the player, including clicking, hovering, etc., and obtains the player log.

(j) The business side periodically feeds back the collected player logs to the data side.

(k), the data side optimizes the group prediction model according to the collected player logs.

The first point that needs to be explained is that since the embodiment of the present invention uses a machine learning model to recommend a team in combination with the player's gameplay preferences and team characteristics, after the player joins a team that has been recommended by the group prediction model, the next step is to stay The rate/weekly retention rate will be better than the original logic, and the online time and active days will also be better than the original logic. Wherein, if the player's online time in one day exceeds a certain threshold, it can be determined that the day is an active day. The second point that needs to be explained is that when displaying the recommended teams on the recommendation panel, the odd-numbered positions can be displayed from the teams recommended by the group prediction model, and the even-numbered positions can be from the teams recommended by the original logic.

The method provided by the embodiment of the present invention realizes team recommendation for the player by using the machine learning model combined with the player's gameplay preference and team characteristics, which not only enriches the recommendation methods, makes the team recommendation more personalized, but also greatly improves the accuracy of the recommendation Not only does it increase the probability of players applying through the team, but also the players are more active after joining the recommended team, and the user experience is higher.

In addition, in the embodiment of the present invention, when recommending teams, in addition to the above-mentioned recommendation methods, team recommendations can also be combined with the original logic, so that teams from two different recommendation modes are cross-displayed on the recommendation panel, which greatly improves The flexibility of team recommendation is more intelligent.

Fig. 11 is a schematic structural diagram of a group recommendation device provided by an embodiment of the present invention. Referring to Figure 11, the device includes:

The first acquiring module 1101 is configured to acquire task preference information of the user to be recommended for each user to be recommended;

A screening module 1102, configured to, among candidate groups in an active state, filter out a group to be predicted that matches the user to be recommended according to the task preference information and the group characteristics of the candidate group;

The second obtaining module 1103 is used to obtain the probability that the user to be recommended joins each group in the group to be predicted;

A sorting module 1104, configured to sort the groups contained in the to-be-predicted group according to the probability value, and obtain the first group recommendation list of the to-be-recommended user;

A recommending module 1105, configured to recommend groups to the user to be recommended based on the first group recommendation list.

In another embodiment, the groups in the first group recommendation list are arranged in descending order according to the probability value, and the recommendation module 1105 is configured to determine in the first group recommendation list the A specified group that has an association relationship with the user to be recommended, the specified group is a group that includes the target user, and the target user is a user that has an association relationship with the user to be recommended; according to the number of target users in the group, the to the smallest order, adjust the order of the groups contained in the specified group; place the specified group after sorting adjustment on the top of the first group recommendation list; based on the sorting adjustment process The first group recommendation list is used to make group recommendations to the user to be recommended.

In another embodiment, the recommending module 1105 is configured to, if the specified group contains at least two groups with the same number of target users, then for each of the at least two groups, obtain the The degree of intimacy between the user to be recommended and the target user included in the group is stated; and the at least two groups are sorted according to the degree of intimacy.

In another embodiment, the screening module 1102 is also used to screen users who are active within the first time period and have not joined any group to obtain the users to be recommended; or, screen the users within the first time period The user who has joined the group but the joined group is an inactive group obtains the user to be recommended; screens the groups that are active within the first time period and has vacancies to obtain the candidate group .

In another embodiment, the second obtaining module 1103 is configured to obtain the probability that the user to be recommended joins each group in the group to be predicted based on the established group prediction model.

In another embodiment, referring to Figure 12, the device further includes:

The training module 1106 is used to use the first user who has joined the group and the group that has joined in the second time period as positive samples; for the second user who is active in the second time period and has not joined any group The user randomly matches at least one group for the second user in the positive sample to obtain a negative sample; obtains the first feature information of the positive sample, and obtains the second feature of the negative sample information; performing model training based on the first feature information and the second feature information to obtain the group prediction model.

In another embodiment, the training module 1106 is further configured to, for each sample in the positive sample, obtain the first user profile information of the first user in the sample; obtain the group information contained in the sample The first group profile information and the first member profile information of the member users in the group; the feature information of the first user profile information, the first group profile information and the first member profile information Crossover processing to obtain first crossover feature information; classify continuous feature information in the first group profile information, the first user profile information, the first member profile information, and the first cross feature information Bucket processing: performing one-bit effective encoding on the feature information after the bucketing process and the discrete feature information except the continuous feature information to obtain the feature information of the sample.

In another embodiment, the training module 1106 is further configured to, for each sample in the negative sample, obtain the second user profile information of the second user in the sample; obtain the group information contained in the sample The second group profile information and the second member profile information of the member users in the group; the feature information of the second user profile information, the second group profile information and the second member profile information Cross process to obtain second cross feature information; classify the second group profile information, the second user profile information, the second member profile information, and the continuous feature information in the second cross feature information Bucket processing: performing one-bit effective encoding on the feature information after the bucketing process and the discrete feature information except the continuous feature information to obtain the feature information of the sample.

In another embodiment, the recommendation module 1105 is configured to obtain a second group recommendation list of the user to be recommended, the second group recommendation list is obtained based on whether the user to be recommended meets the group joining conditions if there are duplicate groups in the first group recommendation list and the second group recommendation list, then filter out the duplicate groups in the second group recommendation list; based on the The first group recommendation list and the second group recommendation list after filtering and deduplication are used to make group recommendations to the user to be recommended.

In another embodiment, the recommendation module 1105 is configured to send the first group recommendation list and the filtered and deduplicated second group recommendation list to the waiting list after obtaining the group recommendation request. Recommending the user's terminal, so that the terminal lists the groups contained in the first group recommendation list and the second group recommendation list on the recommendation interface in a cross-sorted manner according to the arrangement order of the groups exhibit.

In another embodiment, referring to Fig. 13, the device further includes:

A collection module 1107, configured to collect the user log of the user to be recommended, the user log records the exposure of the recommended group and the operations performed by the user to be recommended on the recommended group;

An update module 1108, configured to update the positive samples and negative samples used for the group prediction model training according to the user log;

The training module 1106 is configured to optimize the group prediction model based on the updated positive samples and the updated negative samples.

The device provided by the embodiment of the present invention realizes team recommendation for the player by using the machine learning model combined with the player's gameplay preference and team characteristics, which not only enriches the recommendation methods, makes the team recommendation more personalized, but also greatly improves the accuracy of the recommendation. Not only does it increase the probability of players applying through the team, but also the players are more active after joining the recommended team, and the user experience is higher.

In addition, in the embodiment of the present invention, when recommending teams, in addition to the above-mentioned recommendation methods, team recommendations can also be combined with the original logic, so that teams from two different recommendation modes are cross-displayed on the recommendation panel, which greatly improves The flexibility of team recommendation is more intelligent.

It should be noted that when the group recommendation device provided in the above embodiment performs group recommendation, it only uses the division of the above functional modules as an example for illustration. In practical applications, the above functions can be assigned to different functional modules according to needs. To complete means to divide the internal structure of the device into different functional modules to complete all or part of the functions described above. In addition, the group recommendation device provided in the above embodiments is based on the same concept as the group recommendation method embodiment, and its specific implementation process is detailed in the method embodiment, and will not be repeated here.

Fig. 14 is a schematic structural diagram of a server for group recommendation according to an exemplary embodiment, and the server may be used to implement the group recommendation method shown in any of the foregoing exemplary embodiments. Specifically: referring to FIG. 14, the server 1400 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (Central Processing Unit, CPU) 1422 (for example, one or more processors ) and memory 1432, one or more storage media 1430 (such as one or more mass storage devices) for storing application programs 1442 or data 1444. Wherein, the memory 1432 and the storage medium 1430 may be temporary storage or persistent storage. The program stored in the storage medium 1430 may include one or more modules (not shown in the figure).

The server 1400 can also include one or more power supplies 1428, one or more wired or wireless network interfaces 1450, one or more input and output interfaces 1458, and/or, one or more operating systems 1441, such as Windows Server™, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc. At least one instruction, at least one section of program, code set or instruction set is stored in the memory 1432, and the at least one instruction, the at least one section of program, the code set or instruction set are loaded and executed by the processor to realize the above implementation The group recommendation method described in the example.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, and can also be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. The above-mentioned The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, and the like.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (14)

1. A kind of 1. group recommending method, it is characterised in that the described method includes：

   For each user to be recommended, the task preference information of the acquisition user to be recommended；

   It is special according to the group of the task preference information and the candidate group in the candidate group in active state Sign, filters out and the matched group to be predicted of user to be recommended；

   Obtain the probability that the user to be recommended adds each group in the group to be predicted；

   The group included in the group to be predicted is ranked up according to probable value, obtains first group of the user to be recommended Group recommendation list；

   Group's recommendation is carried out to the user to be recommended based on first group recommendation list.
2. 2. according to the method described in claim 1, it is characterized in that, the group in first group recommendation list is according to probability It is worth descending order arrangement, it is described to be pushed away based on first group recommendation list to user's progress group to be recommended Recommend, including：

   Determine that there are the designated group of incidence relation, the finger with the user to be recommended in first group recommendation list Grouping group is the group comprising targeted customer, and the targeted customer is the user with the user to be recommended there are incidence relation；

   According to the order that targeted customer's quantity in group is descending, the group included in the designated group arrange suitable Sequence adjusts；

   Designated group after sequence adjusts is placed in the top of first group recommendation list；

   Based on first group's recommendation list after sequence adjustment processing, group's recommendation is carried out to the user to be recommended.
3. 3. according to the method described in claim 2, it is characterized in that, the order descending according to targeted customer's quantity, The adjustment that puts in order is carried out to the group included in the designated group, including：

   If comprising at least two groups that targeted customer's quantity is consistent in the designated group, at least two group In each group, obtain the user to be recommended and the cohesion of the targeted customer included in the group；

   At least two group is ranked up according to the cohesion.
4. 4. according to the method described in claim 1, it is characterized in that, the method further includes：

   Screening does not add in active state and the user of any group in first time period, obtains the user to be recommended； Or, the user that screening has added group in the first time period but added group is inactive group, obtain described User to be recommended；

   Screening, in active state and there are the group in room, obtains the candidate group in the first time period.
5. 5. according to the method described in claim 1, it is characterized in that, acquisition user's addition to be recommended is described to be predicted The probability of each group in group, including：

   Based on established group's prediction model, obtain the user to be recommended and add each group in the group to be predicted Probability.
6. 6. according to the method described in claim 5, it is characterized in that, the method further includes：

   Using added in second time period group the first user and added group as positive example sample；

   For being in active state in the second time period and not adding the second user of any group, in the positive example sample At least one group is matched for the second user at random in this, obtains negative example sample；

   The fisrt feature information of the positive example sample is obtained, and obtains the second feature information of the negative example sample；

   Model training is carried out based on the fisrt feature information and the second feature information, obtains group's prediction model.
7. 7. according to the method described in claim 6, it is characterized in that, the fisrt feature information for obtaining the positive example sample, Including：

   For each sample in the positive example sample, the first profile information of the first user in the sample is obtained；

   Obtain the first one-tenth of Member Users in the first group's profile information and the group of the group included in the sample Member's profile information；

   First profile information, first group profile information and first membership profile information are carried out special Reference ceases cross processing, obtains the first cross feature information；

   By first group profile information, first profile information, first membership profile information and described Continuous characteristic information in first cross feature information carries out a point bucket and handles；

   Characteristic information after the processing of undue bucket and the discrete features information in addition to the continuous characteristic information are carried out One efficient coding, obtains the characteristic information of the sample.
8. 8. according to the method described in claim 6, it is characterized in that, the second feature information for obtaining the negative example sample, Including：

   For each sample in the negative example sample, the second user profile information of second user in the sample is obtained；

   Obtain the second one-tenth of Member Users in the second group's profile information and the group of the group included in the sample Member's profile information；

   The second user profile information, second group profile information and second membership profile information are carried out special Reference ceases cross processing, obtains the second cross feature information；

   By second group profile information, the second user profile information, second membership profile information and described Continuous characteristic information in second cross feature information carries out a point bucket and handles；

   Characteristic information after the processing of undue bucket and the discrete features information in addition to the continuous characteristic information are carried out One efficient coding, obtains the characteristic information of the sample.
9. 9. the method according to any claim in claim 1 to 8, it is characterised in that described to be based on described first group Group recommendation list carries out group's recommendation to the user to be recommended, including：

   Second group's recommendation list of the user to be recommended is obtained, second group recommendation list is based on described to be recommended Whether user, which meets group, adds what conditional filtering obtained；

   If there is the group repeated in first group recommendation list and second group recommendation list, described the The group of the repetition is filtered out in two group's recommendation lists；

   Second group's recommendation list based on first group recommendation list and after filtering duplicate removal, to described to be recommended User carries out group's recommendation.
10. 10. the method according to right wants 9, it is characterised in that described to be based on first group recommendation list and process Second group's recommendation list after duplicate removal is filtered, group's recommendation is carried out to the user to be recommended, including：

    After the recommendation request of group is got, by first group recommendation list and second group after filtering duplicate removal Group recommendation list is sent to the terminal of the user to be recommended, so that the terminal putting in order according to group, to intersect row Interface is being recommended by the mode of sequence by the group included in first group recommendation list and second group recommendation list On be shown.
11. 11. the method according to any claim in claim 1 to 8, it is characterised in that the method further includes：

    The user journal of the user to be recommended is collected, the user journal have recorded exposure status and the institute of the group of recommendation State the operation that user to be recommended performs the group of recommendation；

    According to the user journal to for carrying out positive example sample and the progress of negative example sample that group's prediction model is trained Renewal；

    Based on the positive example sample after renewal and the negative example sample after renewal, optimize group's prediction model.
12. 12. a kind of group's recommendation apparatus, it is characterised in that described device includes：

    First acquisition module, for for each user to be recommended, the task preference information of the acquisition user to be recommended；

    Screening module, in the candidate group in active state, according to the task preference information and the candidate The group characteristics of group, filter out and the matched group to be predicted of user to be recommended；

    Second acquisition module, the probability of each group in the group to be predicted is added for obtaining the user to be recommended；

    Sorting module, for being ranked up the group included in the group to be predicted according to probable value, obtains described waiting to push away Recommend first group's recommendation list of user；

    Recommending module, for carrying out group's recommendation to the user to be recommended based on first group recommendation list.
13. A kind of 13. storage medium, it is characterised in that be stored with the storage medium at least one instruction, at least one section of program, Code set or instruction set, at least one instruction, at least one section of program, the code set or the instruction set are by the processing Device is loaded and performed to realize the group recommending method as described in any claim in claim 1 to 11.
14. 14. a kind of server recommended for group, it is characterised in that the server includes processor and memory, described At least one instruction, at least one section of program, code set or instruction set are stored with memory, described at least one instructs, is described At least one section of program, the code set or instruction set is loaded by the processor and performed to realize as in claim 1 to 11 Group recommending method described in any claim.